Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.14721 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911450408157184 |
|---|---|
| author | Xiao, Zikai Tu, Jianhong Zou, Chuhang Zuo, Yuxin Li, Zhi Wang, Peng Yu, Bowen Huang, Fei Lin, Junyang Liu, Zuozhu |
| author_facet | Xiao, Zikai Tu, Jianhong Zou, Chuhang Zuo, Yuxin Li, Zhi Wang, Peng Yu, Bowen Huang, Fei Lin, Junyang Liu, Zuozhu |
| contents | Web agents require massive trajectories to generalize, yet real-world training is constrained by network latency, rate limits, and safety risks. We introduce \textbf{WebWorld} series, the first open-web simulator trained at scale. While existing simulators are restricted to closed environments with thousands of trajectories, WebWorld leverages a scalable data pipeline to train on 1M+ open-web interactions, supporting reasoning, multi-format data, and long-horizon simulations of 30+ steps. For intrinsic evaluation, we introduce WebWorld-Bench with dual metrics spanning nine dimensions, where WebWorld achieves simulation performance comparable to Gemini-3-Pro. For extrinsic evaluation, Qwen3-14B trained on WebWorld-synthesized trajectories improves by +9.2\% on WebArena, reaching performance comparable to GPT-4o. WebWorld enables effective inference-time search, outperforming GPT-5 as a world model. Beyond web simulation, WebWorld exhibits cross-domain generalization to code, GUI, and game environments, providing a replicable recipe for world model construction. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_14721 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | WebWorld: A Large-Scale World Model for Web Agent Training Xiao, Zikai Tu, Jianhong Zou, Chuhang Zuo, Yuxin Li, Zhi Wang, Peng Yu, Bowen Huang, Fei Lin, Junyang Liu, Zuozhu Artificial Intelligence I.2 Web agents require massive trajectories to generalize, yet real-world training is constrained by network latency, rate limits, and safety risks. We introduce \textbf{WebWorld} series, the first open-web simulator trained at scale. While existing simulators are restricted to closed environments with thousands of trajectories, WebWorld leverages a scalable data pipeline to train on 1M+ open-web interactions, supporting reasoning, multi-format data, and long-horizon simulations of 30+ steps. For intrinsic evaluation, we introduce WebWorld-Bench with dual metrics spanning nine dimensions, where WebWorld achieves simulation performance comparable to Gemini-3-Pro. For extrinsic evaluation, Qwen3-14B trained on WebWorld-synthesized trajectories improves by +9.2\% on WebArena, reaching performance comparable to GPT-4o. WebWorld enables effective inference-time search, outperforming GPT-5 as a world model. Beyond web simulation, WebWorld exhibits cross-domain generalization to code, GUI, and game environments, providing a replicable recipe for world model construction. |
| title | WebWorld: A Large-Scale World Model for Web Agent Training |
| topic | Artificial Intelligence I.2 |
| url | https://arxiv.org/abs/2602.14721 |